Classify rustc mechanisms across diagnostic codes

docs/probes/e0308_classify_sites.py

@@ -19,7 +19,7 @@
 NOTE_EXP = re.compile(r"=\s*note:\s*expected (\w+(?: \w+)?) `(.+?)`\s*$")
 NOTE_FOUND = re.compile(r"^\s*found (\w+(?: \w+)?) `(.+?)`\s*$")
 
-def read_blocks(path):
+def read_blocks(path, codes=("E0308",)):
     lines = [ANSI.sub("", l.rstrip("\n")) for l in open(path, encoding="utf-8", errors="replace")]
     blocks, cur = [], None
     for i, l in enumerate(lines):
@@ -33,7 +33,7 @@ def read_blocks(path):
                 blocks.append(cur); cur = None; continue
             cur["lines"].append(l)
     if cur: blocks.append(cur)
-    return [b for b in blocks if b["code"] == "E0308"]
+    return blocks if codes is None else [b for b in blocks if b["code"] in codes]
 
 def block_sites(b):
     """Expand one block into canonical mismatch sites."""
@@ -385,4 +385,5 @@ def main(path, out):
             print("  %s:%s:%s  %s | %s  [%s] callee=%s" %
                   (s["file"], s["line"], s["col"], s["expected"], s["found"], s["delta"], s["callee"]))
 
-main(sys.argv[1], sys.argv[2])
+if __name__ == "__main__":
+    main(sys.argv[1], sys.argv[2])

docs/probes/e0308_partition_2026-08-22.md

@@ -7,6 +7,15 @@ existing 15-category vocabulary of
 
 This is a **measurement**, not a repair. Nothing in this lane changes a single emitted type.
 
+**Cross-code correction (2026-08-23):** this document is an E0308 projection, not a mechanism
+board. The absent-`Clone`-bound mechanism classified here as `A-clone = 12` also produces 6 E0277
+and 4 E0599 manifestations in this same log: its cross-code population is **22**, not 12. Therefore
+the percentages below are shares *within E0308*, never mechanism shares, and the staffing ranking
+derived from those shares is withdrawn. The cross-code instrument, complete TSV, and receipt are
+[`rustc_mechanism_partition_2026-08-23.md`](rustc_mechanism_partition_2026-08-23.md). The 22 is at
+this document's `967b5bc1b92` ref, not the current certified board at `98b18cdc81e`, and must not
+be compared with that board's raw-row total.
+
 **Every cluster below is a CANDIDATE root, not a root.** Mechanical clustering plus shared
 provenance establishes that a set of sites *looks like* one producer decision; only changing one
 modeled decision, rerunning the whole board, and observing the exact site conversions establishes
@@ -306,12 +315,13 @@ width of one file.
 BOX-WRAP) and two more sit in two files each (D, ELEM-COLL), which is the strongest cost-shape
 signal in this run: 49 of 154 sites sit behind six decisions in eight files.
 
-## Cost shape — the staffing question the brief asked
+## Cost shape — E0308 projection only; staffing conclusion withdrawn
 
-The board is **not** a long tail of independent site repairs. Four clusters (R1, R2, D, T3) cover
+Within E0308, the board is **not** a long tail of independent site repairs. Four clusters (R1, R2, D, T3) cover
 88 of 154 = 57%, and six near-single-file clusters cover another 49. The tail is 7 residue sites in
-5 files. This agrees in shape with the 2026-08-21 finding (7 largest categories = 79%) even though
-the population is different.
+5 files. This concentration does **not** rank mechanisms: the cross-code classifier proves that a
+mechanism can extend beyond this denominator, and only `ABSENT_CLONE_BOUND` has been classified
+across codes so far.
 
 | cluster | sites | suggested owner | why that owner |
 |---|---:|---|---|